accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Christopher Tubbs (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ACCUMULO-769) MapReduce API should not use Configuration to set Job state at submission time (ambiguous semantics)
Date Wed, 12 Dec 2012 21:47:21 GMT

    [ https://issues.apache.org/jira/browse/ACCUMULO-769?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13530377#comment-13530377
] 

Christopher Tubbs commented on ACCUMULO-769:
--------------------------------------------

[~billie.rinaldi] - I agree that the API churn is frustrating, but I think if we're going
to support the "new" API, we should do it in a way that is consistent and sensible for users
familiar with this API, and I think this is consistent with the design intent, documentation,
and examples in the "new" API. I think it makes sense to consider ACCUMULO-695 's solution
separately, in the context of the "old" API, with a goal of consistency with that API's design
intent, documentation, and examples.
                
> MapReduce API should not use Configuration to set Job state at submission time (ambiguous
semantics)
> ----------------------------------------------------------------------------------------------------
>
>                 Key: ACCUMULO-769
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-769
>             Project: Accumulo
>          Issue Type: Bug
>          Components: client
>    Affects Versions: 1.4.1, 1.4.0
>            Reporter: Christopher Tubbs
>            Assignee: Christopher Tubbs
>            Priority: Minor
>             Fix For: 1.5.0
>
>
> ACCUMULO-267 made this change, but I think it was the wrong way to go about it.
> From the comments on ACCUMULO-267:
> This is the wrong way to go about doing this fix. The reason why it took a JobContext
is so that it could accept a "Job" object. This was modeled after the pattern Hadoop was using
for FileOutputFormat, which is somewhat the standard for conventions in configuring MR jobs.
> While JobContext does specifically state that's what it's purpose is, it is a base class,
and Job extends JobContext, and includes a comment that describes it as holding the state
of the job at submission time. This API should really be taking a "Job" object, rather than
a "JobContext" object. Further, because Job is the only JobContext that actually works as
intended here, the change from JobContext to Job does not require any deprecation, because
Job will still work, and any other JobContext that isn't a Job will still fail. (We would
have to deprecate the ones that were added in 1.4 that took a Configuration object, though...
because those were never "correct", if we are going off of the conventions set by Hadoop's
provided OutputFormats).
> It is somewhat annoying to deprecate something in 1.5 that was added in 1.4... especially
since it allows people to go back to what they were doing before. But, I think it might be
worth it to be consistent with the established conventions, and to clarify the semantics of
the methods (we are, after all, modifying the state of a job we are about to submit, and not
just an arbitrary configuration, which is used for all sorts of things).

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message